filtering data.frame based on row_number()

Question

UPDATE: dplyr has been updated since this question was asked and now performs as the OP wanted

I´m trying to get the second to the seventh line in a data.frame using dplyr.

I´m doing this:

require(dplyr)
df <- data.frame(id = 1:10, var = runif(10))
df <- df %>% filter(row_number() <= 7, row_number() >= 2)

But this throws an error.

Error in rank(x, ties.method = "first") : 
  argument "x" is missing, with no default

I know i could easily make:

df <- df %>% mutate(rn = row_number()) %>% filter(rn <= 7, rn >= 2)

But I would like to understand why my first try is not working.

I could do that too, but why `df <- df %>% filter(row_number() <= 7, row_number() >= 2)` is wrong? — Daniel Falbel, Sep 23 '14 at 12:04
I don't know the real reason behind that. A double `filter` appears to work. — akrun, Sep 23 '14 at 12:05
It's a bug. Please file an issue on https://github.com/hadley/dplyr/issues — hadley, Sep 23 '14 at 22:53
Is this question still valid, it doesn't seem to give an error anymore, it could be confusing if it isn't flagged as outdated — Jonno Bourne, Nov 16 '16 at 10:38
I think it is useful to have around as long as it is clear, that it is now out of date, this way, people (like me) looking for help can see that this is not a problem. I edited the post for clarity. — Jonno Bourne, Nov 16 '16 at 13:18

score 103 · Accepted Answer · answered Feb 11 '16 at 19:59

103

Actually dplyr's slice function is made for this kind of subsetting:

df %>% slice(2:7)

(I'm a little late to the party but thought I'd add this for future readers)

answered Feb 11 '16 at 19:59

talat

68,970
21
126
157

1

thanks, this was really helpful as the error reoccurred for me. I later found out that this is a inconsistency with how row_number() treats data tables, see: http://stackoverflow.com/questions/23861047/unique-rows-in-dplyr-row-number-from-tbl-dt-inconsistent-to-tbl-df – Alex Jun 21 '16 at 23:10

Spacedman · Answer 2 · 2014-09-23T12:47:27.800

29

The row_number() function does not simply return the row number of each element and so can't be used like you want:

• ‘row_number’: equivalent to ‘rank(ties.method = "first")’

You're not actually saying what you want the row_number of. In your case:

df %>% filter(row_number(id) <= 7, row_number(id) >= 2)

works because id is sorted and so row_number(id) is 1:10. I don't know what row_number() evaluates to in this context, but when called a second time dplyr has run out of things to feed it and you get the equivalent of:

> row_number()
Error in rank(x, ties.method = "first") : 
  argument "x" is missing, with no default

That's your error right there.

Anyway, that's not the way to select rows.

You simply need to subscript df[2:7,], or if you insist on pipes everywhere:

> df %>% "["(.,2:7,)
  id        var
2  2 0.52352994
3  3 0.02994982
4  4 0.90074801
5  5 0.68935493
6  6 0.57012344
7  7 0.01489950

edited Sep 23 '14 at 12:47

answered Sep 23 '14 at 12:37

Spacedman

92,590
12
140
224

12

The purpose of `row_number()` is definitely to return the row number (hence the name!) and this behaviour is a bug. (Also you don't need `.` in your piping example) – hadley Sep 23 '14 at 22:52
Note to anyone thinking `row_number` shows the `row_number` code. It doesn't. You want the `row_number_prototype` C++ function. – Spacedman Sep 24 '14 at 07:01
1

Would you care to explain how the `"["(.,2:7,)` syntax works? It's really interesting solution. – Konrad Aug 04 '16 at 14:49
3

Its just the way almost everything in R can be written as a function. Try `"+"(1,3)`. – Spacedman Aug 04 '16 at 15:12
2

@Konrad as alternative you can write, slightly more readable than the "[" syntax: df %>% .[2:7, ] – Agile Bean Dec 29 '17 at 09:49
1

@Spacedman is correct, row_number does *not* return exactly the row number as vector, see e.g. the output of `datasets::airquality %>% row_number`. If you want to use it as an index vector containing the row number, you must convert it to a numeric vector like `datasets::airquality %>% row_number %>% as.numeric` – Agile Bean Apr 05 '19 at 03:17

score 8 · Answer 3 · answered Feb 11 '16 at 18:56

8

Here is another way to do row-number based filtering in a pipeline.

    df <- data.frame(id = 1:10, var = runif(10))

    df %>% .[2:7,]

    > id     var
      2  2 0.28817
      3  3 0.56672
      4  4 0.96610
      5  5 0.74772
      6  6 0.75091
      7  7 0.05165

answered Feb 11 '16 at 18:56

dabsingh

281
1
3
10

1

It's slower than `slice`, but it does't drop `NA` (e.g. `df %>% .[c(NA,2,4,7),]`) which could be useful in some cases. – Bastien Jan 09 '18 at 12:53

score 0 · Answer 4 · answered Jan 13 '23 at 08:31

Another option using subset:

df <- data.frame(id = 1:10, var = runif(10))
subset(df, row.names(df) %in% 2:7)
#>   id        var
#> 2  2 0.75924106
#> 3  3 0.17096427
#> 4  4 0.10886090
#> 5  5 0.98703882
#> 6  6 0.04190195
#> 7  7 0.73268672

^{Created on 2023-01-13 with reprex v2.0.2}

filtering data.frame based on row_number()

4 Answers4

Linked